Abstract: In the context of AI exploration of criminal justice, the AI conviction prediction accelerates the exclusion of traditional concepts. From the practice of AI conviction prediction, the accurate prediction function is the focus and presents a distinctive feasibility. The traditional conviction theory system and its operational mechanism have been subverted as a whole, and the AI conviction prediction has begun to foster a self-contained knowledge system. The precondition for the practice of AI conviction prediction is judicial big data and its judicial conviction logic. The intelligent decomposition of judicial big data lays the objectivity, authenticity and reliability of AI prediction function. The precise prediction function of AI conviction has successive development forms such as class push and intelligent case push. It establishes its complete prediction mechanism through algorithm logic and rules to form knowledge pedigree and modeling, and configures a complete verification mechanism. In the operation mode of AI conviction prediction, the forecasting of confession and punishment cases has unique advantages and opportunities.